Problems about make_visualizations.py

[Ljiaqii]

Hi! Thank you for sharing your great work! When I was studying make_ visualizations. py, I met some problems. You define activations_ hook on line 21 of the program, but line 22 self.gradients = grads is not really initialized. So I found 36 lines of self activations_ hook will not call the function of line 21, which leads to errors in subsequent programs. Therefore, do you have a good way to solve this problem or can you provide the latest version of the code? Thank you very much and look forward to your reply! Best regards!

[teddykoker]

Hi!

This is the latest version of my code for doing GradCAM. activations_hook() should be called when .backword() is called on the output predictions (see docs).

If you could provide a stack trace of the error that would be very helpful!

[Ljiaqii]

Hi! Thank you for your quick reply! About the problem of .backward () is proved that I carelessly omitted it, I'm very sorry about it and I have corrected it. Moreover, I intend to realize 3D image classification visualization based on your code (the cube of my program is a three-dimensional gray image with the size of 32 × 32× 32(z,y,x)). But now I don't quite understand a few lines of code. Could you please tell me how to modify it?

Here are the codes :

def grad_cam_cube(model, test_data_loader): model = GradCAMResnet(model) model.eval()

for batch_idx, (inputs, targets, feat) in enumerate(test_data_loader):
    inputs, targets = inputs.cuda(), targets.cuda()    #  batchsize = 1, input [1,1,32,32,32]

    inputs, targets = Variable(inputs, requires_grad=False), Variable(targets)
    outputs = model(inputs)     #outputs:tensor([[ 4.2578,-4.4285]], device='cuda:0', grad_fn=<AddmmBackward>)
    outputs1 = outputs[0, 0]     #outputs1:tensor(4.2578, device='cuda:0', grad_fn=<SelectBackward>)
    outputs2 = outputs[0, 1]     #outputs2:tensor(-4.4285, device='cuda:0', grad_fn=<SelectBackward>)
    outputs[0,0].backward()      # otherwise outputs[0,1].backward()
    gradients, activations = model.gradients, model.activations  # gradients [1,2048,1,1,1]    # What does 2048 mean?
    pooled_gradients = torch.mean(gradients, dim=[0, 2, 3])
    for i in range(activations.shape[1]):
        activations[:, i, :, :] *= pooled_gradients[i]

    heatmap = torch.mean(activations, dim=1).squeeze()     #heatmap: tensor(0.0021, device='cuda:0')
    heatmap = torch.relu(heatmap)
    heatmap /= torch.max(heatmap)     #heatmap={Tensor} tensor(0.0021, device='cuda:0'  #Does the heatmap need to return 32 × 32 × 32？

The following code is my definition of GradCAMResnet:

class GradCAMResnet(torch.nn.Module): def init(self, model):
super().init() self.model = model

def activations_hook(self, grads):
    self.gradients = grads

def _forward_impl(self, x):
    # See note [TorchScript super()]
    x = self.model.conv1(x)  
    x = self.model.bn1(x)
    x = self.model.relu(x)
    x = self.model.maxpool(x)

    x = self.model.layer1(x)
    x = self.model.layer2(x)
    x = self.model.layer3(x)
    x = self.model.layer4(x)
   
    #I added the following two lines of code.
    self.activations = x.detach()  
    x.register_hook(self.activations_hook)  
    x = self.model.avgpool(x)
    x = torch.flatten(x, 1)
    x = self.model.fc(x)

    return x

def forward(self, x):
    return self._forward_impl(x)

Thank you for your help! Looking forward to your reply!